Company context for coding agents.

Picture a normal workday.

A customer reports that authentication fails after a session expires. Support records what the customer saw. Someone researches earlier reports. The team discusses the cause and decides not to store payment credentials. A coding agent is asked to fix the bug.

The last agent may only receive four words: “Fix that authentication bug.” It does not know what was tried, what the team decided, or what the fix must avoid. A person has to explain the whole story again.

People are still the bridge.

People copy messages from one tool to another. They rewrite notes, repeat decisions, and warn the next agent about failed attempts. This works when the work is small. It breaks when many tasks move at once.

The problem is not that teams lack more transcripts. The problem is that no shared record follows the work from its first report to its final result.

The useful context includes the reason.

An agent needs to know the current goal, the decisions and limits, what was tried, what evidence supports the current view, what result is expected, and why the work matters to the company or customer.

Old conversations still matter as evidence. They should support the current state of the work, not replace it.

The next agent should start with a clear briefing.

Termyte is being built to connect the related activity, keep the state current, and give the next agent a short explanation it can use. The goal is not to send every message to every agent. The goal is to send the right work context at the right time.

This is a direction, not a finished promise.

Today, Termyte is a local open-source context engine for coding agents. It does not include company-source integrations or connectors.

The wider company context layer still needs to be built and proven. The coding-agent alpha is the first narrow product on that path.